The problem that comes before administration
A state that wants to tax a harvest must first know what a field is, whose it is, and how big. A state that wants to conscript must know who exists, where they sleep, and how old they are. A state that wants to plan a road, a vaccination campaign or a famine response must be able to move a description of a place from that place to a desk somewhere else without the description falling apart on the way. None of this is possible against reality as it is actually lived, which is local, plural, seasonal, negotiated and full of exceptions that everyone involved understands and nobody has written down.
The precondition, then, is a conversion. Practices that were legible to the people performing them have to be turned into a representation that is uniform across places, portable across distances, and comparable across cases. James C. Scott named this property legibility and made it the organising concept of Seeing Like a State, arguing that the modern state’s characteristic projects — cadastral mapping, permanent surnames, standardised measures, freehold tenure, planned settlement, scientific forestry — are best understood as attempts to make a complex society simple enough to be administered [1].
Two clarifications matter before going further, because the argument is regularly flattened in both directions.
First, this is a thesis, not a finding. It is an influential interpretive framework, widely taught and widely applied, and it has serious critics whose objections are taken up later in this article. Treating it as established fact does it no favours.
Second, the thesis is not a story of progress and not a story of decline. Simplification is not a pathology of bad states; it is the operating requirement of any state, including ones doing things most people want done. The analytically interesting questions are which simplification was chosen, what it excluded, who was made visible to whom, and what happened when the abstraction and the ground disagreed.
The founding act: fixing a boundary
If legibility has a founding instrument, it is the cadastral survey. Kain and Baigent trace rural property mapping in Europe and the European colonies from the Renaissance to the nineteenth century, with roots reaching back to Roman practice, and treat the resulting maps not as neutral cartography but as political instruments — the means by which land-based national power was consolidated and extended across the Netherlands, France, England, the Nordic countries, the German lands, the Habsburg territories and overseas possessions [5].
The mechanism is worth stating precisely, because it is the template for everything that follows. A survey does not record a boundary. It decides one. Before the survey there is a set of overlapping, time-varying claims: this family has cropped the upper slope for three generations; that lineage grazes it after harvest; the strip by the stream is used by whoever needs it in a dry year. After the survey there is a line, and the line has two sides. The instrument, the chain and the mark are not describing a pre-existing fact; they are manufacturing one, and the manufactured fact is the only version the tax office, the court and the mortgage lender will ever see.
This is why the survey is the founding act rather than merely the first act. It creates the object — the parcel — on which every later administrative operation depends. Everything downstream, from valuation to inheritance to compulsory purchase, is operating on that object and not on the slope.
Identifying the person: surnames, numbers, measures
The same logic applied to land was applied to people, and by the same means: fix an identifier, make it permanent, and require it in dealings with the state.
Scott, Tehranian and Mathias examine the permanent family surname as exactly this kind of instrument. Their argument is that state naming practices and local customary naming practices differ sharply in purpose: customary names are contextual, plural and sufficient among people who already know one another, while state naming is designed to identify a unique individual or household unambiguously to a stranger holding a list [2]. A patronym that changes each generation is perfectly serviceable in a village and useless in a tax roll that must be matched across decades. The permanent inherited surname is the fix, and it was frequently imposed rather than adopted.
House numbering is the same instrument pointed at the dwelling rather than the person, and its history makes the purpose unusually explicit. Tantner argues against the intuitive assumption that house numbers were introduced to help people find their way, locating their origin instead in the overlap of early modern police, military and fiscal administration: numbering made it possible to billet soldiers, to assess and reach household resources, to run conscription, and to keep watch on people the authorities considered dangerous. He tracks the spread across eighteenth-century Europe — Prussia from 1737, Madrid around 1750, Vienna in 1770, Paris in 1779 — and notes both the variety of systems tried and the popular resistance the practice met before it became invisible infrastructure [6].
Standard measures complete the set. A unit that varies by valley, by commodity and by custom is not an obstacle to trade so much as an obstacle to comparison from a distance: it prevents a central authority from ranking, aggregating or auditing. The modern endpoint of that project is a unit system defined by fixed numerical values of physical constants rather than by any artefact or local practice, maintained internationally and published as a single reference [7]. Read as administrative history rather than as metrology, the significance of a universal unit is that it makes a measurement taken by a stranger in one place binding on a decision taken by a stranger in another.
Analytically, these three instruments — permanent name, fixed address, common unit — are one instrument with three targets. Each converts a locally sufficient practice into a globally comparable key. Each is usually described in the language of convenience and is, in origin, a technology of identification.
Counting, and the categories counting requires
A census cannot count people. It can only count people-of-some-kind, and the kinds have to exist before the enumeration begins. This is not a defect in census design; it is a structural feature of counting anything. But it has a consequence that is easy to state and hard to absorb: the act of counting participates in creating the category counted.
Ian Hacking made this the centre of his account of “making up people”, arguing that classifications of persons interact with the people classified, who then change, which in turn alters the classification — the looping effect. Among the engines he lists as driving the process, the first three are counting, quantifying and creating norms, followed by correlating, medicalising, biologising, geneticising, normalising, bureaucratising, and finally resistance by the classified themselves [8]. Counting is not the observation stage that precedes the political stage. It is one of the engines.
A recent and fully documented instance: in March 2024 the United States revised its federal standards for collecting race and ethnicity data. The revision replaced the separate race and ethnicity questions with a single combined question, added a Middle Eastern or North African category for people who had previously been classified within the White category, and replaced the tabulation category “Two or More Races” with “Multiracial and/or Multiethnic” [9]. No population moved. No one’s ancestry changed. What changed was the set of boxes, and with it every downstream number: population shares, disparity measures, the denominators of health statistics, the evidentiary base for civil-rights enforcement, and the demographic descriptions people encounter and use to describe themselves.
The analytical point is not that categories are arbitrary. Some fit lived identity better than others, which is precisely why they are fought over. The point is that a category is a decision with statistical consequences, and that a state which counts is not standing outside the thing it counts.
The register makes property
Land registration is where the abstraction bites hardest, because a register does not merely record claims — it converts them. Customary tenure typically consists of layered, conditional and overlapping entitlements: a household cultivates, a lineage allocates, an in-law retains a residual claim, a widow keeps use rights, seasonal users cross. A title is singular. Registration therefore requires that a bundle of relationships be resolved into an owner, and that resolution is a distributive act with beneficiaries and losers.
International policy has partially absorbed this. The Voluntary Guidelines on the Responsible Governance of Tenure, endorsed by the Committee on World Food Security in May 2012, explicitly extend responsible governance to “all forms of tenure: public, private, communal, indigenous, customary and informal”, which is a recognition in principle that legitimate rights exist outside any register [10]. Practice is more mixed than the principle.
The empirical record does not support a single verdict, and this is one of the places where the honest answer is that outcomes depend on design and context. Ali, Deininger and Goldstein evaluated Rwanda’s land tenure regularisation pilots using a geographic discontinuity design and report improved land access for legally married women — about 76 percent of married couples — better recording of inheritance rights without gender bias, large investment impacts concentrated in soil conservation and particularly pronounced for female-headed households, and a reduction rather than an increase in land market activity, which lets them rule out distress sales [11]. On these measures the register protected people who had been vulnerable without it.
A study of Zambian informal settlements points the other way on a different margin. Bwalya Umar and colleagues surveyed all 302 households that had obtained title deeds in two settlements in Lusaka and Chongwe districts and found that formalisation gave secure recognition to people who had lived under constant threat of eviction — while curtailing the claims of secondary claimants such as extended family members, in favour of a man, a spouse and biological children. Roughly half of respondents reported no change in family conflict, at least a third reported increased trust between spouses, about half reported no change in women’s decision-making authority and 42 percent reported an increase. The authors’ own summary is cautionary rather than triumphal: what is being formalised is a pre-existing distribution of power, and titling can legalise it as easily as reform it [12].
Set side by side, these two studies say something more useful than either alone. A register is not intrinsically extractive or intrinsically protective. It is a mechanism for making one version of a claim official, and its distributive result depends on whose version the enumerator writes down.
The double edge
The extraction reading of legibility is the famous one, and it is only half the ledger. The same apparatus that lets a state find you for taxation and conscription is what lets you be found for entitlement, protection and care.
Consider the civil register. The World Health Organization states that roughly two-thirds of deaths worldwide are never registered and therefore never enter vital statistics, and that the births of tens of millions of children go unregistered every year [13]. Those are not merely gaps in a dataset. An unregistered death is a cause of death that never informs a health system’s understanding of what is killing its population. An unregistered birth is a child who may be unable to prove age, nationality or parentage, with consequences for schooling, protection from early marriage, inheritance and freedom of movement. The United Nations Legal Identity Agenda, established in 2018, exists to coordinate work toward Sustainable Development Goal target 16.9 — legal identity for all, including birth registration, by 2030 — measured by the share of children under five whose births are registered, and it notes that statelessness alone affects more than ten million people [14].
This is the reason legibility cannot be treated as simply a mode of domination. To be uncounted is not to be free; it is frequently to be unprotected. A person with no registered birth, no address and no recognised claim to the land they farm is not outside the state’s reach when the state wants something. They are only outside its reach when they want something.
The honest formulation is that legibility is a capability, and capabilities are directional only in use. The same register that identifies a household for a levy identifies it for a cash transfer. The same address system that enables surveillance enables an ambulance. The interesting variable is not how much a state can see but what it is obliged to do with what it sees, and who can contest the record.
Where the thesis is contested
Scott’s framework has drawn substantial criticism, and the disagreement is worth characterising rather than adjudicating.
Tania Murray Li’s critique is the most direct. She proposes five moves beyond Scott’s analytic scheme: dropping the spatial picture of an “up there”, all-seeing state that spreads outward into non-state space; recognising that many parties beyond the state govern, including reformers, scientists and non-governmental agencies; widening the frame from authoritarian high modernism to the broader rationality of “improvement” aimed at the welfare of populations; taking seriously the local knowledge and practice that sits outside planning; and, most pointedly, replacing the question of why schemes fail with the question of what schemes actually do, given their messy, contradictory and conjunctural effects [3]. The force of this is that “the state” in the legibility story can be a unified intending actor in a way that no real administration is, and that failure-versus-success is the wrong axis for effects that are neither.
Mara Loveman’s work supplies a second line of objection, this time empirical. In 1851 Brazil’s Decree 798 made civil registration of births and deaths obligatory. From the first days of January 1852, men, women and children across the impoverished northeastern backlands convened in settlements and towns to protest the decree; local authorities reported abandoning their posts; disturbances were reported in at least thirty-one localities with crowds estimated at one hundred to several thousand people, and troops were deployed. On 29 January 1852 the decree was suspended. Civil registration was effectively cancelled, and so was the first general census of the Brazilian Empire, which had been scheduled for July of that year. Brazil would not attempt another general census for two decades, and obligatory civil registration was not implemented with any measure of success until the country became a republic in 1889 [4]. Loveman also records that land surveys, censuses and the standardisation of weights and measures were recurrent triggers of popular revolt, and that in Brazil a separate uprising specifically targeted standard weights.
The structural claim in that episode is the important one. A legibility project is not something a state does to a population; it is something a state attempts, and the attempt can be defeated. Local actors are not merely the material on which simplification operates. They evade, misreport, negotiate, capture the process for their own ends, and occasionally stop it outright for forty years.
A defender of the thesis can absorb much of this — Scott’s book is largely a catalogue of schemes that failed, and mētis, the local practical knowledge that planning cannot encode, is his own term. The unresolved disagreement is about weighting and about the unit of analysis: whether the useful protagonist is a centralised state with a coherent optical project, or a shifting field of actors — colonial officials, missionaries, scientists, development agencies, local elites — pursuing improvement with partial and conflicting instruments. That dispute is open, and this article does not resolve it.
The failure mode that keeps returning
The characteristic failure of legibility is not simplification. It is forgetting that a simplification was made — treating the register as the territory and then making policy on the abstraction.
The mechanism is easy to see once named. A representation is built for one purpose, with known exclusions. It becomes the only version visible at the centre. Later decisions are taken by people who never saw the original, using a description whose exclusions are no longer documented. Interventions are then optimised against the description, and where the description and the ground disagree, the ground is treated as the error. Scott’s own case studies — scientific forestry, planned settlement, agricultural modernisation — are variations on this sequence [1].
The sequence did not end with paper. Amnesty International’s investigation into the Dutch childcare benefits scandal found that the tax authority’s automated system for detecting benefit fraud incorporated nationality as a risk factor, and concluded that this resulted in discrimination and racial profiling [15]. As a legibility problem, the structure is exactly the classical one: a proxy variable stood in for a thing that could not be measured directly, the proxy was optimised against, and the people the proxy misdescribed had no practical way to contest the description before it produced consequences. What differs from the cadastral case is speed and scale, not logic.
This suggests a diagnostic that generalises. The dangerous configuration is not a state with a large register. It is a register with three properties at once: the simplification is invisible to its users; there is no low-friction route by which a misdescribed person can correct the record; and the abstraction is coupled directly to a consequential decision without a human in a position to notice that the case in front of them does not fit.
Forecast, assumptions and what would falsify it
The following is a prediction, clearly separated from the historical and empirical material above.
Horizon: through 2032.
Prediction: the volume of state legibility will continue to increase across most jurisdictions, principally through digital identity systems, interoperable registries and administrative-data linkage; the contested political question will shift from whether the state can see to who may correct the record and on what terms. Concretely, I expect correction, appeal and contestability provisions to become a more prominent axis of identity-system disputes than data collection itself.
Assumptions: that no widespread rollback of digital identity programmes occurs; that the coverage gaps documented in vital registration continue to attract multilateral funding [13, 14]; and that automated decision systems continue to be attached to existing administrative registers rather than replaced by new collection.
Observable indicators, in order of usefulness: statutory rights of rectification and appeal appearing in identity and registry legislation; published error, exclusion and correction-latency rates for national identity and benefit systems; litigation and regulatory enforcement centred on wrongly recorded status rather than on collection consent; and the appearance of registry categories that carry an explicit provisional or contested designation.
Disconfirmation condition: the prediction should be treated as wrong if, by 2032, the dominant pattern of disputes in this area remains focused on the legality of collection and retention, with contestability and correction provisions not materially more prominent in legislation, litigation or published system metrics than they are now.
The underlying claim is modest and, I think, robust across the disagreement between Scott and his critics. Whether one reads the register as the optical instrument of a centralising state or as the contested product of many actors pursuing improvement, the same practical question decides how it lands on any particular person: when the record and the person disagree, which one is treated as the error, and how expensive is it for the person to be believed. The survey mark does not know what it decided. The sheet on the plane table was blank until someone drew on it. Everything that follows depends on remembering that a choice was made there at all.
